The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for refreshing its analysis index by running npx gitnexus or npm install -g gitnexus. These commands involve fetching and executing the vendor's CLI tool from the public NPM registry.
[COMMAND_EXECUTION]: The skill uses a specialized Node.js script (scripts/evidence-provenance.mjs) to execute git for repository metadata retrieval and python3 to perform atomic, race-resistant file moves via low-level libc system calls. It also attempts to execute a project-local runner (.gitnexus/run.cjs) if present.
[PROMPT_INJECTION]: The skill processes untrusted repository files, which constitutes a potential surface for indirect prompt injection.
Ingestion points: The agent reads source code files, GitNexus graph outputs, and existing plan documents from the repository.
Boundary markers: The skill utilizes a structured ledger to track investigation steps and tags all findings in the final plan with explicit evidence classes ([verified], [graph], [inferred], [assumed]).
Capability inventory: The skill possesses capabilities to read arbitrary repository files, execute specific system tools (git, python3), and write plan documents to a defined repository path (docs/plans/).
Sanitization: While the instructions mandate rigorous source verification of graph-derived claims, they do not specify explicit sanitization or filtering of the content read from repository files.

gitnexus-plan